The current forecast date is 25 February, showing data until 05 March.
The main forecasting model is an unweighted ensemble of three individual models:
covid19.nhs.data)The time series ensemble and regression models are trained on the last 6 weeks of data from 14 January; the case-convolution model is trained on the last 9 weeks of data (from 24 December), of which 6 are used in the likelihood. Forecasts are made for the following 14 days.
For details of the fitted models, see our preprint “Comparative assessment of methods for short-term forecasts of COVID-19 admissions in England at the local level”.
Note: for clarity, we only show forecasts for Trusts which have admitted five or more patients in the last 7 days.
We derive forecasts of national and regional hospital admissions from the Trust-level admissions. For each of the individual models, the forecast samples are summed to get national/regional forecasts, then summarised into quantile forecasts, and then the ensemble forecast created. These are compared below to forecasts made using EpiNow2.
We compare last weeks forecasts (from 18 February 2022) to the current observed data.
We derive forecasts of national hospital admissions from the Trust-level admissions. For each of the individual models, the forecast samples are summed to get national forecasts, then summarised into quantile forecasts. The ensemble forecast is created from the national quantile forecasts.
We derive forecasts of regional hospital admissions from the Trust-level admissions. For each of the individual models, the forecast samples are summed to get regional forecasts, then summarised into quantile forecasts. The ensemble forecast is created from the regional quantile forecasts.